Removes the need to keep the spin matrices in memory for N > 5. Instead adds a function called add_dipolar_field!(op::Array{ComplexF64, 2}, B::Vec3 which adds B ⋅ S to op by constructing the relevant matrix elements of the spin matrices on the fly. For N=6 this is marginally faster than keeping the matrices stored (3-5%). I also tried N=10, where it was about 10% faster than storing the matrices. So it seems like this is a reasonable approach to dealing with the N > 5 case, at least for now.
I added a test that ensures this function does what it should. I also added an N=6 case to the spin rescaling tests, so that the N > 5 case is at least run during routine testing.
Removes the need to keep the spin matrices in memory for N > 5. Instead adds a function called
add_dipolar_field!(op::Array{ComplexF64, 2}, B::Vec3which addsB ⋅ Stoopby constructing the relevant matrix elements of the spin matrices on the fly. For N=6 this is marginally faster than keeping the matrices stored (3-5%). I also tried N=10, where it was about 10% faster than storing the matrices. So it seems like this is a reasonable approach to dealing with the N > 5 case, at least for now.I added a test that ensures this function does what it should. I also added an
N=6case to the spin rescaling tests, so that theN > 5case is at least run during routine testing.